Automatic document filing according to user categories

Suppose you have just acquired a number of articles into a personal digital library. The classification procedures described here would allow those articles to be automatically sorted in your pre-existing desktop folders. We use the Chi-square statistic to find keywords that we use to classify the articles, and the Support Vector Classifier to organize the articles in folders. Moreover we have adapted the process so that minimal feedback from users should improve classification results, and folders do not contain overly many articles for information management convenience. We achieved an average of 96% accuracy over 10 trials with the Reuters 21578 news data, and 89% accuracy over 10 trials with biomedical literature downloaded from the PubMed digital library. We conclude that in a personal desktop environment, whether general or highly specific, our classification methods for automatic filing into the user’s own familiar folders would achieve high accuracy.